Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin

Anders Sandneseng H{\o}gden; Asgeir S{\o}rensen; Dong Trong Nguyen; Emir Cem Gezer; Mael Korentin Ivan Moreau; Roger Skjetne

REVIEW 3 major objections 3 minor 1 cited by

Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin

T0 review · 3 major / 3 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper's central claim is that a digital-physical testbed—model-scale vessels, layered simulators, sensor simulation, and a digital twin—can carry autonomous-ship algorithms from cost-effective simulation to physical basin trials, and…

desk verdict Open, modular testbed with genuinely useful artifacts; the quantitative fidelity claims are thinner than the language, but the package deserves a serious look. read the letter →

arxiv 2505.06787 v5 pith:YMOFOJLC submitted 2025-05-10 cs.RO cs.SYeess.SY

Emir Cem Gezer , Mael Korentin Ivan Moreau , Anders Sandneseng H{o}gden , Dong Trong Nguyen , Roger Skjetne , Asgeir S{o}rensen This is my paper · ORCID

classification cs.ROcs.SYeess.SY

keywords MaritimeAutonomousSurfaceShips(MASS)SimulationTestingValidationModelDigitalTwinDigitalizationHydrodynamicmodeling

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper describes an in-house digital-physical testbed intended to support development, testing, verification, and validation of Maritime Autonomous Surface Ship (MASS) algorithms. The testbed combines a fleet of small-scale model vessels, a laboratory wave basin, a reduced-order simulator, a high-fidelity hydrodynamic simulator, a sensor-rich robotics simulator, and a 3D digital twin, all sharing a common software interface. The central demonstration is a four-corner stationkeeping maneuver on the model vessel C/S Voyager: the physical vessel followed the reference with 0.180 m position RMSE, while the reduced-order simulator achieved 0.062 m, which the authors present as evidence that the simulation-to-physical pipeline is operational. A sympathetic reader would care because the testbed is meant to be an open, reusable benchmark platform, with hydrodynamic datasets and software released publicly, lowering the barrier for other laboratories to conduct model-scale autonomy experiments.

What carries the argument

The argument is carried by a layered simulation hierarchy connected to a shared software and hardware stack. Hydrodynamic coefficients are produced by boundary-element potential-flow analyses of approximate half-hull meshes and feed a high-fidelity simulator; a reduced-order simulator treats each vessel as a uniform rectangular prism with diagonal added-mass and damping terms (20 percent of rigid-body mass in surge, 100 percent in sway and roll) and integrates the 6DOF rigid-body equations of motion with a fourth-order Runge–Kutta step. The reduced-order simulator and the physical vessels expose the same interface, and the physical fleet runs a common onboard stack with an extended Kalman filter observer, a proportional-integral velocity controller, a proportional-derivative pose controller, a control-Lyapunov/control-barrier-function thrust allocator, and motion-capture-based pose feedback. The four-corner stationkeeping test on C/S Voyager is the concrete demonstration that a controller can be developed in the reduced-order model and executed in the basin.

What would settle it

Run a bollard-pull test on C/S Voyager and compare the measured thrust-versus-command curve with the reduced-order simulator's assumed thruster scaling; if the measured curve differs substantially, the 0.180 m physical and 0.062 m simulated RMSEs would be a coincidental match rather than a validated model.

Watch

Extended reading notes

Core claim

The paper claims that a single integrated pipeline, running from low-fidelity simulation up to physical model-scale testing, can support the full design and validation loop for autonomous ship guidance, navigation, and control. The evidence is a four-corner stationkeeping test executed with the same pose controller in both the reduced-order simulator and the physical basin vessel: the simulator produced a position RMSE of 0.062 m, a yaw RMSE of 0.439 degrees, and a velocity RMSE of 0.031 m/s, while the physical vessel produced 0.180 m, 4.295 degrees, and 0.036 m/s. On the authors' terms, this shows that an algorithm developed in the simulator can be transferred to the physical testbed with the same control stack, though the gap between simulated and physical tracking error also indicates where model fidelity remains incomplete.

Load-bearing premise

The load-bearing premise is that the simulator coefficients reflect the real vessels: they come from potential-flow analyses of approximate half-hull meshes with hand-set added-mass and damping scale factors (20 percent surge, 100 percent sway and roll), and the paper states that no bollard-pull test has been run, so the thruster force model is uncalibrated; if those coefficients are wrong, the claimed simulation-to-physical correspondence is not established.

Editorial extensions

If this is right

An algorithm developed in the reduced-order simulator can be run on a physical model vessel with the same software stack, making iteration between simulation and experiment inexpensive.
Publicly released hydrodynamic datasets and open-source simulators let other laboratories reconstruct the same vessels digitally, making the testbed reproducible rather than a one-off setup.
The same pipeline can in principle be extended from model scale to semi-full-scale ferries and full-scale research vessels, reusing the architecture rather than starting from scratch.
The standardized hardware and software architecture means new model vessels can be added to the fleet with minimal re-engineering, so the testbed's capability grows incrementally.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The 0.062 m versus 0.180 m RMSE gap suggests the reduced-order model's thruster and wave-disturbance terms, not the control algorithm, dominate the remaining error; a bollard-pull calibration plus added wave excitation would be a direct test of that reading.
The shared interface between the reduced-order simulator and the physical vessel makes the testbed a natural benchmark for sim-to-real transfer in maritime reinforcement learning, where the same policy could be trained in fast simulation and evaluated in the basin.
Because the testbed publishes its hydrodynamic data, cross-laboratory comparisons of autonomy algorithms become possible, provided the basin conditions and motion-capture ground truth are also standardized.
Forward-speed hydrodynamic data, which the paper says it plans to add, would extend the pipeline's validity from stationkeeping to transit maneuvers, where damping and wave encounter effects are stronger.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Quick take: this is an infrastructure paper, and judged as infrastructure it mostly works. The genuinely new thing is the integrated package — open simulators (mcsimpy, shoeboxpy), publicly released hydrodynamic datasets, a shared ROS 2 control stack, and a digital twin — all tied to a physical fleet in a wave basin. That combination gives the community something reusable rather than another one-off demo. Credit where due: the control laws are standard but the stability argument is correct, and the four-corner run on C/S Voyager shows the whole pipeline, from reduced-order model to physical vessel, actually executes. The paper is also honest about its limits; the missing bollard-pull test and the approximate WAMIT meshes are stated outright, not buried.

Soft spots, in proportion. The main one is that the sim-to-physical fidelity claim rests on essentially one demonstration: a single four-corner run per platform, with no error bars, no repeated trials, and no uncertainty quantification. The reported 0.062 m vs 0.180 m RMSE difference could easily shift with run-to-run variability, initial conditions, or unmodeled basin effects. The paper's own language — "transition seamlessly," "facilitates verification and validation" — overshoots what a single uncalibrated run can support. Also, the hydrodynamic coefficients come from simplified half-hull meshes and hand-set added-mass/damping scale factors; that is fine for a progress report, but it means the sim fidelity is asserted rather than demonstrated. These are not fatal flaws; they are gaps between evidence and language.

Who this is for: anyone building similar lab infrastructure, or anyone wanting a ready-made benchmark for MASS algorithms, will get real value from the released artifacts. The paper itself is more of a capability statement than a scientific advance. I would want to see repeated trials, a bollard-pull calibration, and at least one external comparison before treating the testbed as a validated benchmark. But the package is open, modular, and the authors are explicit about what is missing.

Recommendation: send it to peer review. It is not a breakthrough, but it is a solid, honest infrastructure contribution that the community can build on, and the weaknesses are addressable through standard experimental practice rather than conceptual error.

Referee Report

3 major / 3 minor

Summary. The manuscript describes an open, modular digital-physical testbed at NTNU's Marine Cybernetics Laboratory for developing and validating MASS autonomy algorithms. It comprises a fleet of small-scale C/S vessels, a high-fidelity simulation environment (mcsimpy) with hydrodynamic data from WAMIT/VERES, a reduced-order simulator (shoeboxpy), a Stonefish sensor simulator, and a Unity digital twin with remote-control capabilities. The software stack is ROS 2 based and includes thrust allocation, velocity- and pose-control modules, and an EKF observer. The central demonstration is a four-corner stationkeeping test of C/S Voyager in the reduced-order simulator and in the physical basin, reporting position/yaw/velocity RMSEs of 0.062 m/0.439 deg/0.031 m/s in simulation and 0.180 m/4.295 deg/0.036 m/s in the basin. The authors claim that the testbed facilitates a seamless simulation-to-physical transition and supports verification and validation of autonomy algorithms.

Significance. The main strength is the infrastructure itself: the software and hydrodynamic datasets are openly released with DOIs, making this a potentially reusable community benchmark and lowering the entry barrier for maritime autonomy testing. The stability analysis of the velocity controller (Eqs. 5-8) is standard and correct. The four-corner experiment demonstrates that the integrated pipeline - observer, controller, thrust allocation, and actuator drivers - operates end to end in both simulation and the physical basin. However, the quantitative fidelity claims rest on a single uncalibrated four-corner run, and the hydrodynamic and thruster models are acknowledged to be approximate. As a functional demonstration the paper is convincing; as a validation of simulation-to-physical fidelity it is not yet supported.

major comments (3)

[3.3.2, Fig. 4] Only one four-corner run is reported for each platform, with no repeated trials, error bars, or uncertainty quantification. The reported difference between simulated (0.062 m) and physical (0.180 m) position RMSE cannot be separated from run-to-run variability, initial-condition sensitivity, or unmodeled basin effects. Because the paper uses this result to claim that the testbed facilitates verification and validation and allows seamless transition, the demonstration needs either replicate runs with summary statistics or a correspondingly explicit caveat that the result is a functional check rather than a fidelity validation.
[3.1.2, 3.4, 5] The fidelity chain is not quantitatively established. Section 3.4 states that WAMIT meshes are approximate half-hull meshes with geometric simplifications; Section 3.1.2 sets added-mass and damping scale factors by hand (20% surge, 100% sway and roll); and the final paragraph of Section 5 states that no bollard-pull test has been conducted, leaving the thruster model uncalibrated. These are clearly disclosed, but they imply that the four-corner comparison cannot quantify how well the simulators reproduce the physical vessel's dynamics. The paper should add a sensitivity analysis or a calibration experiment (e.g., bollard-pull) before claiming quantitative simulation-to-physical correspondence; alternatively, the fidelity claims should be explicitly limited to qualitative or functional correspondence.
[3.1.1, Fig. 4] The demonstration uses only the reduced-order simulator; the high-fidelity WAMIT-based mcsimpy environment is not exercised or validated in the presented experiment. The paper's full design-and-verification-pipeline claim therefore remains unsubstantiated for the high-fidelity layer. A simulation-to-simulation comparison or a high-fidelity four-corner run would close this gap, or the claim should be scoped to the reduced-order layer.

minor comments (3)

[3.3.2] In the text below Eq. (6), 'feedforward compansation' is a typo for 'feedforward compensation'.
[Abstract] The pipeline direction is described inconsistently: the abstract states that the pipeline 'starts from low-fidelity and moves up to high-fidelity' while the article text says it 'starts with high-fidelity simulation models'; please harmonize these statements.
[3.1.1] The software name appears as 'mcsimpy2' in Section 3.1.1 and as 'mcsimpy' elsewhere; please use one consistent name.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the testbed demonstration is an independent empirical comparison, and the hydrodynamic and scale-factor inputs are not fitted to the reported results.

full rationale

The paper's central demonstration is a four-corner stationkeeping run of the C/S Voyager, with RMSE values reported separately for the reduced-order simulator (0.062 m position) and the physical basin experiment (0.180 m). The simulator's hydrodynamic coefficients are stated as coming from WAMIT runs on approximate half-hull meshes, and the reduced-order model uses explicitly declared hand-set added-mass and damping scale factors (20% surge, 100% sway and roll). These are model inputs; the paper does not fit them to the four-corner data or to the physical trajectory before calling the result a prediction. The comparison is therefore an externally checkable output, not an input disguised as a result. The paper also candidly states in Section 5 that no bollard-pull test has been conducted and that small-scale models only approximate full-scale behavior, which limits the strength of the fidelity claims but is not a circularity. Self-citations (e.g., Gezer et al. 2024 for thrust allocation, Skjetne et al. 2017 for the four-corner test) are contextual references to prior methods and do not carry the load-bearing argument; no uniqueness theorem or ansatz is imported from the authors' prior work. Accordingly, no circular step is exhibited.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No new physical entities, mediators, or conserved quantities are introduced. The digital twin and simulators are representations, not invented entities. The main assumptions are standard marine hydrodynamics models plus the unvalidated fidelity of WAMIT-based coefficients and hand-set scaling factors.

free parameters (3)

Surge added-mass scale factor = 0.20 (20% of rigid-body mass)
Chosen ad hoc in the reduced-order model to approximate added mass in surge; not identified from experiments or data; affects Eq. (3) dynamics. Section 3.1.2.
Sway and roll added-mass scale factor = 1.00 (100% of rigid-body mass)
Hand-set default for sway and roll added mass in the simplified model; no empirical identification. Section 3.1.2.
Linear damping scale factors = not stated (predefined factors)
Diagonal linear damping D is estimated by scaling rigid-body values with predefined factors; the factors and their justification are not given. Section 3.1.2.

assumptions (5)

domain assumption Six-DOF rigid-body equations of motion with Coriolis, added-mass, damping, and restoring terms (Eq. 3) describe the vessel dynamics
Standard marine craft model from Fossen (2021); assumed valid for the model vessels in the testbed.
domain assumption JONSWAP and Pierson-Moskowitz spectra represent the wave environment in the high-fidelity simulator
Standard ocean engineering spectral models; used to generate sea states in mcsimpy. Section 3.1.1.
domain assumption WAMIT potential-flow results on simplified half-hull meshes provide accurate hydrodynamic coefficients
The high-fidelity simulators and released datasets rely on WAMIT outputs from approximate 3D scans with geometric simplifications and no experimental validation. Section 3.4.
standard math Krasovskii-LaSalle invariance principle
Used to conclude uniform global asymptotic stability of the PI-RFF velocity controller from negative semi-definite V-dot. Section 3.3.2.
domain assumption Model-scale behavior in the basin is representative of full-scale vessel behavior for early-stage validation
The whole pipeline assumes transferability from model-scale to full scale; the paper itself notes small-scale models do not fully capture full-scale behavior. Section 5.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin." pith.science (2026). https://pith.science/paper/YMOFOJLC

@misc{pith2026250506787,
  author       = {Pith},
  title        = {Pith review of: Digital-physical testbed for ship autonomy studies in the Marine Cybernetics Laboratory basin},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YMOFOJLC}},
  note         = {Machine review of arXiv:2505.06787}
}

read the original abstract

The algorithms developed for Maritime Autonomous Surface Ships (MASS) are often challenging to test on actual vessels due to high operational costs and safety considerations. Simulations offer a cost-effective alternative and eliminate risks, but they may not accurately represent real-world dynamics for the given tasks. Utilizing small-scale model ships and robotic vessels in conjunction with a laboratory basin provides an accessible testing environment for the early stages of validation processes. However, designing and developing a model vessel for a single test can be costly and cumbersome, and researchers often lack access to such infrastructure. To address these challenges and enable streamlined testing, we have developed an in-house testbed that facilitates the development, testing, verification, and validation of MASS algorithms in a digital-physical laboratory. This infrastructure includes a set of small-scale model vessels, a simulation environment for each vessel, a comprehensive testbed environment, and a digital twin in Unity. With this, we aim to establish a full design and verification pipeline that starts from low-fidelity and moves up to high-fidelity simulation models of each vessel, and thereby to the model-scale testing of the vessel in the laboratory basin. Further advancement allows moving towards semi-full-scale validation with R/V milliAmpere1 and full-scale validation with R/V Gunnerus. In this work, we present our progress on the development of this testbed environment and its components, demonstrating its effectiveness in enabling ship autonomy guidance, navigation, and control (GNC) algorithms.

Figures

Figures reproduced from arXiv: 2505.06787 by the authors.

**Figure 2.** C/S Voyager (on the left) and C/S Enterprise I (on the right) in the Stonefish [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Control architecture of the testbed environment showing the closed-loop sys [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: : Four corner tests for C/S Voyager in simulation (left) and physical basin [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

**Figure 5.** Figure 5: : Shared hardware architecture across the C/S fleet. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

pacSTL: PAC-Bounded Signal Temporal Logic from Data-Driven Reachability Analysis
cs.LO 2025-11 conditional novelty 6.0 of 10

pacSTL composes PAC-bounded reachable sets with interval STL to compute spec-level robustness intervals that contain an unseen trajectory's robustness with probability ≥ 1−ε.

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages · cited by 1 Pith paper

[1]

Thruster-assisted position mooring of C/S Inocean Cat I drillship

Bjørnø, Jon (2016). “Thruster-assisted position mooring of C/S Inocean Cat I drillship”. MA thesis. url: http://hdl.handle.net/11250/2402895. Brekke, Edmund F et al. (2022). “milliAmpere: An autonomous ferry prototype”. In: Journal of Physics: Conference Series . Vol

work page 2016
[2]

Stonefish: An advanced open-source simulation tool designed for marine robotics, with a ros interface

Cie´ slak, Patryk (2019). “Stonefish: An advanced open-source simulation tool designed for marine robotics, with a ros interface”. In: OCEANS 2019-Marseille . Department of Marine Technology (2025). Marine Cybernetics Laboratory (MC-lab) . https://www.ntnu.edu/imt/lab/cybernetics. Accessed: 2025-03-30. Eide, Egil et al. (2025). “The autonomous urban passe...

work page 2019
[4]

An overview of the marine systems simulator (MSS): A simulink toolbox for marine control systems

12 NTNU (2025). Norwegian Ocean Technology Centre. Accessed: 2025-04-01. url: https: //www.ntnu.edu/norwegian-ocean-technology-centre . Perez, Tristan et al. (2006). “An overview of the marine systems simulator (MSS): A simulink toolbox for marine control systems”. In:Modeling, identification and Control 27.4, pp. 259–275. Pettersen, Kristin Y et al. (199...

work page 2025
[8]

Development of a DP system for CS Enterprise I with Voith Schneider thrusters

SINTEF (2025). ShipX: Hydrodynamic Analysis Software . Accessed: 2025-03-30. url: https://www.sintef.no/en/software/shipx/. Sk˚ atun, H˚ akon Nødset (2011). “Development of a DP system for CS Enterprise I with Voith Schneider thrusters”. MA thesis. url: http : / / hdl . handle . net / 11250 / 237980. Skjetne, Roger (2005). “The maneuvering problem”. PhD t...

work page 2025
[11]

Autonomous docking using direct optimal control

Martinsen, Andreas B et al. (2019). “Autonomous docking using direct optimal control”. In: IFAC-PapersOnLine 52.21. Mo, Harald et al. (2023). “Real-time compensation of residual loads in dynamic position- ing control systems”. MA thesis. url: https://hdl.handle.net/11250/3093878. Moore, T. et al. (2014). “A Generalized Extended Kalman Filter Implementatio...

work page 2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Thruster-assisted position mooring of C/S Inocean Cat I drillship

Bjørnø, Jon (2016). “Thruster-assisted position mooring of C/S Inocean Cat I drillship”. MA thesis. url: http://hdl.handle.net/11250/2402895. Brekke, Edmund F et al. (2022). “milliAmpere: An autonomous ferry prototype”. In: Journal of Physics: Conference Series . Vol

work page 2016

[2] [2]

Stonefish: An advanced open-source simulation tool designed for marine robotics, with a ros interface

Cie´ slak, Patryk (2019). “Stonefish: An advanced open-source simulation tool designed for marine robotics, with a ros interface”. In: OCEANS 2019-Marseille . Department of Marine Technology (2025). Marine Cybernetics Laboratory (MC-lab) . https://www.ntnu.edu/imt/lab/cybernetics. Accessed: 2025-03-30. Eide, Egil et al. (2025). “The autonomous urban passe...

work page 2019

[3] [4]

An overview of the marine systems simulator (MSS): A simulink toolbox for marine control systems

12 NTNU (2025). Norwegian Ocean Technology Centre. Accessed: 2025-04-01. url: https: //www.ntnu.edu/norwegian-ocean-technology-centre . Perez, Tristan et al. (2006). “An overview of the marine systems simulator (MSS): A simulink toolbox for marine control systems”. In:Modeling, identification and Control 27.4, pp. 259–275. Pettersen, Kristin Y et al. (199...

work page 2025

[4] [8]

Development of a DP system for CS Enterprise I with Voith Schneider thrusters

SINTEF (2025). ShipX: Hydrodynamic Analysis Software . Accessed: 2025-03-30. url: https://www.sintef.no/en/software/shipx/. Sk˚ atun, H˚ akon Nødset (2011). “Development of a DP system for CS Enterprise I with Voith Schneider thrusters”. MA thesis. url: http : / / hdl . handle . net / 11250 / 237980. Skjetne, Roger (2005). “The maneuvering problem”. PhD t...

work page 2025

[5] [11]

Autonomous docking using direct optimal control

Martinsen, Andreas B et al. (2019). “Autonomous docking using direct optimal control”. In: IFAC-PapersOnLine 52.21. Mo, Harald et al. (2023). “Real-time compensation of residual loads in dynamic position- ing control systems”. MA thesis. url: https://hdl.handle.net/11250/3093878. Moore, T. et al. (2014). “A Generalized Extended Kalman Filter Implementatio...

work page 2019